Intelligent automatic interpretation of active marine sonar
This dissertation explores the problems raised by the design and construction of a real-time sonar interpreter operating in a three dimensional marine context, and then focusses on two major research issues inherent in sonar interpretation: the treatment of observer and object motion, and the efficient exploitation of the specularity of acoustic reflection. The theoretical results derived in these areas have been tested where appropriate by computer simulation. In the context of mobile marine robotics, the registration of sensory data obtained from differing viewpoints is of paramount importance. Small marine vehicles of the type considered here do not carry sophisticated navigational equipment, and cannot be held stationary in the water for any length of time. The viewpoint registration problem is defined and analysed in terms of the new problem of motion resolution: the task of resolving the apparent motion of objects into that part due to the movement of the observer and that due to the objects' proper motion. Two solutions to this under constrained problem are presented. The first presupposes that the observer orientation is known ~ priori so that only the translational observer motion must be determined. It is applicable to two and three-dimensional situations. The second solution determines both the translational and the rotational motion of the observer, but is restricted to a two-dimensional situation. Both solutions are based on target extensively tested in two tracking techniques, and have dimensions by computer simulation. been The necessary extensions to deal with full three-dimensional motion are also discussed. The second major research issue addressed in this thesis is the efficient use of specularity. Specular echoes have a high intrinsic information content because of the alignment conditions necessary for their generation. In the marine acoustic context they provide a significant proportion of the information available from an acoustic ranger. I suggest a new method that uses directly the information present in specular reflections and the history of the vehicle motion to classify the specular echo sources and infer the local structure of the objects bearing them. The method builds on the output of a motion resolution system. Six distinct types of specular echo source are described and three properties useful for their discrimination are discussed. A suitable inference system for the analysis and classification of specular echo sources is also proposed.